[Question] Support for torch 2.5

[Galaxy-Husky]

Checklist

• 1. I have searched related issues but cannot get the expected help.
• 2. The bug has not been fixed in the latest version.
• 3. Please note that if the bug-related issue you submitted lacks corresponding environment info and a minimal reproducible demo, it will be challenging for us to reproduce and resolve the issue, reducing the likelihood of receiving feedback.

Describe the bug

Hi,
I noticed that the restriction on torch version has been relaxed to 2.5.1:

lmdeploy/requirements/runtime_cuda.txt

Line 18 in dfeee42

torch<=2.5.1,>=2.0.0

But when I tried installing the latest code via: pip install "git+https://github.com/InternLM/lmdeploy.git@main" -U , it would remove torch 2.5.1 and install torch 2.4.0.

Can you tell me how to install lmdeploy without changing the torch version?

Reproduction

pip install "git+https://github.com/InternLM/lmdeploy.git@main" -U

Environment

sys.platform: linux
Python: 3.12.8 | packaged by conda-forge | (main, Dec  5 2024, 14:24:40) [GCC 13.3.0]
CUDA available: True
MUSA available: False
numpy_random_seed: 2147483648
GPU 0,1,2,3: NVIDIA A100-SXM4-40GB
CUDA_HOME: None
GCC: gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
PyTorch: 2.5.1+cu124
PyTorch compiling details: PyTorch built with:
  - GCC 9.3
  - C++ Version: 201703
  - Intel(R) oneAPI Math Kernel Library Version 2024.2-Product Build 20240605 for Intel(R) 64 architecture applications
  - Intel(R) MKL-DNN v3.5.3 (Git Hash 66f0cb9eb66affd2da3bf5f8d897376f04aae6af)
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - LAPACK is enabled (usually provided by MKL)
  - NNPACK is enabled
  - CPU capability usage: AVX2
  - CUDA Runtime 12.4
  - NVCC architecture flags: -gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_90,code=sm_90
  - CuDNN 90.1
  - Magma 2.6.1
  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=12.4, CUDNN_VERSION=9.1.0, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -D_GLIBCXX_USE_CXX11_ABI=0 -fabi-version=11 -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOROCTRACER -DLIBKINETO_NOXPUPTI=ON -DUSE_FBGEMM -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-unused-parameter -Wno-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=old-style-cast -Wno-missing-braces -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, TORCH_VERSION=2.5.1, USE_CUDA=ON, USE_CUDNN=ON, USE_CUSPARSELT=1, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_GLOO=ON, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=1, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF, 

TorchVision: 0.19.0+cu121
LMDeploy: 0.6.4+3009f92
transformers: 4.47.0
gradio: Not Found
fastapi: 0.115.6
pydantic: 2.10.3
triton: 3.1.0
NVIDIA Topology: 
	�[4mGPU0	GPU1	GPU2	GPU3	CPU Affinity	NUMA Affinity�[0m
GPU0	 X 	NV4	NV4	NV4	0-31,64-95	0
GPU1	NV4	 X 	NV4	NV4	0-31,64-95	0
GPU2	NV4	NV4	 X 	NV4	32-63,96-127	1
GPU3	NV4	NV4	NV4	 X 	32-63,96-127	1

Legend:

  X    = Self
  SYS  = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
  NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
  PHB  = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
  PXB  = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
  PIX  = Connection traversing at most a single PCIe bridge
  NV#  = Connection traversing a bonded set of # NVLinks

Error traceback

No response

[RunningLeon]

@Galaxy-Husky hi, sorry for the trouble. Because the requirements for torchvision is not updated to torch2.5.1, so it would install torch2.4.0.
This issue has been fixed in pr #2912 . Pls try on the latest main branch.
requirements/runtime_cuda.txt

- torchvision<=0.19.0,>=0.15.0
+ torchvision<=0.20.1,>=0.15.0

[Galaxy-Husky]

@Galaxy-Husky hi, sorry for the trouble. Because the requirements for torchvision is not updated to torch2.5.1, so it would install torch2.4.0. This issue has been fixed in pr #2912 . Pls try on the latest main branch. requirements/runtime_cuda.txt

- torchvision<=0.19.0,>=0.15.0
+ torchvision<=0.20.1,>=0.15.0

Thanks for your reply.
I tried on the latest main branch before I submit this issue. I'm guessing the previous failure was caused by me not updating torchvision==0.19.0 to 0.20.1.
But it still failed after I tried again by creating a new environment. It would remove torch 2.5.1 and install torch 2.4.1. The following are my commands:

conda create -n lmdeploy python=3.12
pip3 install torch==2.5.1 torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124
pip install "git+https://github.com/InternLM/lmdeploy.git@main" -U 

Could you help me check again?

[RunningLeon]

@Galaxy-Husky ok, let me check again if there still has other problems.

[RunningLeon]

@Galaxy-Husky hi, can you try again install with --use-deprecated=legacy-resolver to pip.

pip install "git+https://github.com/InternLM/lmdeploy.git@main" -U --use-deprecated=legacy-resolver --extra-index-url https://download.pytorch.org/whl/cu124

[Galaxy-Husky]

@RunningLeon Hi, It raised an error:

...
Successfully built lmdeploy
Installing collected packages: triton, lmdeploy
  Attempting uninstall: triton
    Found existing installation: triton 3.1.0
    Uninstalling triton-3.1.0:
      Successfully uninstalled triton-3.1.0
  Attempting uninstall: lmdeploy
    Found existing installation: lmdeploy 0.6.4
    Uninstalling lmdeploy-0.6.4:
      Successfully uninstalled lmdeploy-0.6.4
ERROR: pip's legacy dependency resolver does not consider dependency conflicts when selecting packages. This behaviour is the source of the following dependency conflicts.
torch 2.5.1+cu124 requires triton==3.1.0; platform_system == "Linux" and platform_machine == "x86_64" and python_version < "3.13", but you'll have triton 3.0.0 which is incompatible.
Successfully installed lmdeploy-0.6.4 triton-3.0.0

[RunningLeon]

hi, it seems lmdeploy with torch2.5.1 has been installed successfully. The msg can be ignored since triton version in lmdeploy is fixed to triton==3.0.0.

[Galaxy-Husky]

@RunningLeon I see. BTW, it would be great if you could tell me whether torch2.5.1 work well with triton==3.0.0 or not. And when will lmdeploy upgrade triton to 3.1.0?

[RunningLeon]

hi, triton3.0.0 can work well with torch2.5.1. As for upgrading triton to 3.1.0, there would need a full scale test.

@RunningLeon I see. BTW, it would be great if you could tell me whether torch2.5.1 work well with triton==3.0.0 or not. And when will lmdeploy upgrade triton to 3.1.0?